[reference model] Add shape operators
- fixed up reshape conformance tests to use shape input instead of attribute
- fixed up tile conformance tests to use shape input instead of attribute
- fixed output and output rank of dim op
- allow rank 0 and rank 1 tensors for tosa.shape values (for shape = {})
- added initialization of rank 0 const_shape tensors (for shape = {})
- Update conformance tests to use new rescale attributes
Signed-off-by: Tai Ly <tai.ly@arm.com>
Signed-off-by: Won Jeon <won.jeon@arm.com>
Change-Id: I6cce0d2a9ab066fe20a2abf9d2cfde3eb3d8c18b
diff --git a/verif/generator/tosa_test_gen.py b/verif/generator/tosa_test_gen.py
index b1c53f5..67ac367 100644
--- a/verif/generator/tosa_test_gen.py
+++ b/verif/generator/tosa_test_gen.py
@@ -263,7 +263,7 @@
return gtu.vect_f32_to_bf16(rand_f32)
elif dtype == DType.BOOL:
return self.rng.choice([False, True])
- elif dtype == DType.INT48:
+ elif dtype == DType.INT48 or dtype == DType.SHAPE:
# Special size
return np.int64(self.rng.integers(low, high, size=1))[0]
@@ -1556,15 +1556,24 @@
def build_reshape(
self, op, inputs, args_dict, validator_fcns=None, error_name=None, qinfo=None
):
- assert len(inputs) == 1
+ assert len(inputs) == 2
a = inputs[0]
- new_shape = args_dict["new_shape"]
+ # second input is not properly generated yet
+ # new_shape = inputs[1]
+
+ # modify inputs[1] by a shape tensor from new_shape arg value
+ new_shape_attr = args_dict["new_shape"]
+ shape_array = np.array(new_shape_attr)
+ shape = shape_array.shape
+ new_shape = self.ser.addPlaceholder(shape, DType.SHAPE, shape_array)
+ inputs[1] = new_shape
+
result_tensor = OutputShaper.reshapeOp(
- self.ser, self.rng, a, new_shape, error_name
+ self.ser, self.rng, a, new_shape_attr, error_name
)
# Invalidate Input/Output list for error if checks.
- input_list = [a.name]
+ input_list = [a.name, new_shape.name]
output_list = [result_tensor.name]
pCount, cCount = op["operands"]
num_operands = pCount + cCount
@@ -1588,10 +1597,7 @@
):
return None
- attr = ts.TosaSerializerAttribute()
- attr.ReshapeAttribute(new_shape)
-
- self.ser.addOperator(op["op"], input_list, output_list, attr)
+ self.ser.addOperator(op["op"], input_list, output_list)
compliance = self.tensorComplianceMetaData(
op, a.dtype, args_dict, result_tensor, error_name
@@ -1717,16 +1723,24 @@
def build_tile(
self, op, inputs, args_dict, validator_fcns=None, error_name=None, qinfo=None
):
- assert len(inputs) == 1
+ assert len(inputs) == 2
a = inputs[0]
- multiples = args_dict["multiples"]
+ # second input is not properly generated yet
+ # multiples = inputs[1]
+
+ # modify inputs[1] by a shape tensor from multiples arg value
+ multiples_attr = args_dict["multiples"]
+ shape_array = np.int64(np.array(multiples_attr))
+ shape = shape_array.shape
+ multiples = self.ser.addPlaceholder(shape, DType.SHAPE, shape_array)
+ inputs[1] = multiples
result_tensor = OutputShaper.tileOp(
- self.ser, self.rng, a, multiples, error_name
+ self.ser, self.rng, a, multiples_attr, error_name
)
# Invalidate Input/Output list for error if checks.
- input_list = [a.name]
+ input_list = [a.name, multiples.name]
output_list = [result_tensor.name]
pCount, cCount = op["operands"]
num_operands = pCount + cCount
@@ -1751,10 +1765,7 @@
):
return None
- attr = ts.TosaSerializerAttribute()
- attr.TileAttribute(multiples)
-
- self.ser.addOperator(op["op"], input_list, output_list, attr)
+ self.ser.addOperator(op["op"], input_list, output_list)
compliance = self.tensorComplianceMetaData(
op, a.dtype, args_dict, result_tensor, error_name
@@ -1989,12 +2000,16 @@
in_type_width = self.typeWidth(val.dtype)
out_type_width = self.typeWidth(out_dtype)
+ input_unsigned = False
+ output_unsigned = False
+
if val.dtype == DType.INT8:
input_zp = self.randInt(-128, 128)
in_type_width += 1
elif val.dtype == DType.UINT8:
input_zp = self.randInt(0, 256)
in_type_width += 1
+ input_unsigned = True
elif error_name in [
ErrorIf.InputZeroPointNotZero,
ErrorIf.U16InputZeroPointNotValid,
@@ -2007,6 +2022,7 @@
# Must come after ErrorIf.U16InputZeroPointNotValid check
input_zp = self.rng.choice([0, 32768])
in_type_width += 1
+ input_unsigned = True
else:
input_zp = 0
@@ -2016,6 +2032,7 @@
elif out_dtype == DType.UINT8:
output_zp = self.randInt(0, 256)
out_type_width += 1
+ output_unsigned = True
elif error_name in [
ErrorIf.OutputZeroPointNotZero,
ErrorIf.U16OutputZeroPointNotValid,
@@ -2028,6 +2045,7 @@
# Must come after ErrorIf.U16OutputZeroPointNotValid check
output_zp = self.rng.choice([0, 32768])
out_type_width += 1
+ output_unsigned = True
else:
output_zp = 0
@@ -2116,6 +2134,8 @@
scale32,
double_round,
per_channel,
+ input_unsigned,
+ output_unsigned,
)
self.ser.addOperator(op["op"], input_list, output_list, attr)
@@ -4212,7 +4232,7 @@
},
"reshape": {
"op": Op.RESHAPE,
- "operands": (1, 0),
+ "operands": (2, 0),
"build_fcn": (
build_reshape,
TosaTensorGen.tgBasic,
@@ -4277,7 +4297,7 @@
},
"tile": {
"op": Op.TILE,
- "operands": (1, 0),
+ "operands": (2, 0),
"rank": (1, 6),
"build_fcn": (
build_tile,
@@ -5141,7 +5161,7 @@
@staticmethod
def dimOp(ser, rng, a, axis, error_name=None):
- output_shape = []
+ output_shape = [1]
if error_name == ErrorIf.WrongOutputType:
all_dtypes = [